Drug Library Screen Identifies Inhibitors of Toxic Astrogliosis Ruturaj

Home , Astrogliosis

medRxiv preprint doi: https://doi.org/10.1101/2020.09.15.20195016; this version posted September 18, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available for use under a CC0 license.

1 Drug library screen identifies inhibitors of toxic astrogliosis

2 Ruturaj Masvekar1, Peter Kosa1, Christopher Barbour1, Joshua Milstein1 and Bibiana Bielekova1*

4 1National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda,

5 MD.

6 *To whom correspondence should be addressed: Bibiana Bielekova, MD, Neuroimmunological

7 Diseases Section (NDS), National Institute of Allergy and Infectious Diseases (NIAID), National

8 Institutes of Health (NIH), Building 10, Room 5N248, 10 Center Drive, MSC1444, Bethesda,

9 Maryland 20892, USA. ([email protected]).

11 Abstract

12 Objective: Multiple sclerosis is a chronic neuroinflammatory disorder, in which activated

13 immune cells directly or indirectly induce demyelination and axonal degradation. Inflammatory

14 stimuli also change the phenotype of astrocytes, making them neurotoxic. The resulting ‘toxic

15 astrocyte’ phenotype has been observed in animal models of neuroinflammation and in multiple

16 sclerosis lesions. Proteins secreted by toxic astrocytes are elevated in the cerebrospinal fluid of

17 multiple sclerosis patients and reproducibly correlate with the rates of accumulation of

18 neurological disability and brain atrophy. This suggests a pathogenic role for neurotoxic

19 astrocytes in multiple sclerosis.

20 Methods: Here, we applied a commercially available library of small molecules that are either

21 Food and Drug Administration-approved or in clinical development to an in vitro model of toxic

22 astrogliosis to identify drugs and signaling pathways that inhibit inflammatory transformation of

23 astrocytes to a neurotoxic phenotype.

NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. medRxiv preprint doi: https://doi.org/10.1101/2020.09.15.20195016; this version posted September 18, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available for use under a CC0 license.

24 Results: Inhibitors of three pathways related to the endoplasmic reticulum stress: 1) proteasome,

25 2) heat shock protein 90 and 3) mammalian target of rapamycin reproducibly decreased

26 inflammation-induced conversion of astrocytes to toxic phenotype. Dantrolene, an anti-spasticity

27 drug that inhibits calcium release through ryanodine receptors expressed in the endoplasmic

28 reticulum of central nervous system cells, also exerted inhibitory effect at in vivo achievable

29 concentrations. Finally, we established cerebrospinal fluid SERPINA3 as a relevant

30 pharmacodynamic marker for inhibiting toxic astrocytes in clinical trials.

31 Interpretation: Drug library screening provides mechanistic insight into the generation of toxic

32 astrocytes and identifies candidates for immediate proof-of-principle clinical trial(s).

47 Abbreviations

48 MS - multiple sclerosis

49 CNS - central nervous system

50 RRMS - relapsing-remitting MS

51 PMS – progressive MS

52 CSF - cerebrospinal fluid

53 FDA - Food and Drug Administration

54 PBMCs - peripheral blood mononuclear cells

55 PBS - phosphate-buffered saline

56 LPS – lipopolysaccharide

57 TNFα - tumor necrosis factor α

58 IL1α - interleukin 1 α

59 C3 - complement component 3

60 SERPINA3 - serine protease inhibitor family A member 3

61 EDSS - Expanded Disability Status Scale

62 CombiWISE - Combinatorial Weight-Adjusted Disability Scale

63 MRI - magnetic resonance imaging

64 COMRIS-CTD - Composite MRI scale of CNS tissue destruction

65 MSSS - Multiple Sclerosis Severity Score

66 ARMSS - Age Related Multiple Sclerosis Severity

67 MSDSS - Multiple Sclerosis Disease Severity Scale

68 SP-MS - secondary progressive MS

69 PP-MS - primary progressive MS

70 HD - healthy donors

71 SOMAscan - Slow Off-rate Modified Aptamers scan

72 DMSO - dimethyl sulfoxide

73 MTT - 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide

74 ANOVA - Analysis of Variance

75 FDR - False Discovery Rate

76 CXCL - C-X-C motif chemokine ligand

77 MMP - matrix metalloproteinases

78 CCL - C-C motif chemokine ligand

79 CX3CL1 - C-X3-C motif chemokine ligand 1

80 TNFAIP6 - TNFα induced protein 6

81 CF - complement factors

82 IL1RL1 - interleukin 1 receptor like 1

83 ER - endoplasmic reticulum

84 UPR - unfolder protein response

85 HSP90 - heat shock protein 90

86 p38 MAPK - p38 mitogen-activated protein kinase

87 PI3K - phosphoinositide 3-kinases

88 mTOR - mammalian target of rapamycin

89 S1P - sphingosine 1-phosphate

90 RyR - ryanodine receptor

91 PARP - poly ADP ribose polymerase

92 CYPs - cytochrome P450s

93 COX – cyclooxygenase

94 LTR - leukotriene receptor

95 IC50 - 50% inhibitory concentrations

96 BBB - blood-brain barrier

97 GSK-3β - glycogen synthase kinase-3β

98 TMT - trimethyltin chloride

99 DMF - dimethyl fumarate

100 iPSCs - induced pluripotent stem cell

101 UPS - ubiquitin proteasome system

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116 Introduction

117 Multiple sclerosis (MS) is a chronic inflammatory disease of the central nervous system (CNS),

118 where activated immune cells directly or indirectly contribute to the loss of myelin sheath from

119 CNS axons, leading to neurodegeneration. Affecting over 2.0 million individuals worldwide, MS

120 is the most common non-traumatic neurological disorder in young adults (1).

121 Recruitment of immune cells from blood forms acute MS lesions in relapsing-remitting MS (RR-

122 MS) (2,3). Development of focal lesions diminishes in later stages of MS (progressive MS [P-

123 MS]), even though P-MS patients have levels of immune cell-specific cerebrospinal fluid (CSF)

124 biomarkers indistinguishable from that of RR-MS (4), indicating persistent CNS inflammation.

125 Consequently, the age-related decrease in the efficacy of immunomodulatory drugs on MS

126 disability progression (5) has been attributed to inflammation compartmentalized to CNS tissue,

127 largely inaccessible to current drugs. Alternatively, neurodegenerative mechanisms may drive

128 CNS tissue destruction in P-MS. Inflammation-induced change in the phenotype and function of

129 astrocytes towards “neurotoxic” (or “A1”) astrocytes identified in animal models of

130 neuroinflammation represent a candidate neurodegenerative mechanism in MS (6–8).

131 Immunohistochemistry and in situ hybridization on post-mortem MS brain tissues showed the

132 presence of A1 astrocytes in acute and chronic MS lesions (9).

133 Identification of CNS cell-specific protein clusters within the MS CSF proteome measured by a

134 DNA-aptamer-based platform (i.e., Somascan®, Somalogic Boulder, CO, USA) detected only

135 two clusters of proteins that reproducibly correlated with the rate of accumulation of disability

136 and CNS tissue damage in the independent validation cohort of MS patients (10). One protein

137 cluster was enriched for microglia-secreted proteins while the second cluster constituted proteins

138 secreted mostly by inflammatory stimuli-activated astrocytes. Indeed, proteins of this cluster

139 partially overlap with the “toxic astrocyte” signature identified by Liddelow et al. (9). Because

140 this unbiased screen of CNS cell-enriched protein clusters supported a potential pathogenic role

141 of toxic astrocytes in MS progression, we sought to develop an in vitro model of inflammation-

142 induced neurotoxic astrocyte formation for a drug library screen, with two related aims: 1) to

143 elucidate signaling pathways that underlie toxic astrocyte transformation; and 2) to identify Food

144 and Drug Administration (FDA)-approved drug(s) with a reasonable toxicity profile for

145 immediate use in proof-of-principle clinical trial of toxic astrocyte inhibition in MS.

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162 Methods

163 Peripheral blood mononuclear cells (PBMCs) isolation and stimulation

164 PBMCs were isolated using density gradient centrifugation as described (10). PBMC (1x106

165 cells/ml) were cultured in serum-free X-VIVO (Lonza), and either left untreated (unstimulated)

166 or stimulated with lipopolysaccharide (LPS; Sigma-Aldrich, St. Louis, MO; 100 ng/ml) and

167 CD3/CD28 microbeads (Invitrogen, Carlsbad, CA; at 1:1 beads to cells ratio) to activate

168 simultaneously cells of innate and adaptive immunity. After 24 hours, cell-free supernatants were

169 collected, aliquoted and stored at -80°C until further use. We validated that stimulated PBMC

170 supernatants contained high levels of tumor necrosis factor α (TNFα) and interleukin 1 α (IL1α)

171 using ELISA (R&D Systems, Minneapolis, MN; data not shown).

172

173 Astrocyte cultures and treatments

174 Primary human astrocytes from cerebral cortex (ScienCell, Carlsbad, CA; Catalog# 1800;

175 purchased on 03/2018, 06/2018 and 03/2019) were cultured (105 cells/ml) as per manufacturer’s

176 instructions. After 24 hours, cells were treated with 50% volume/volume of either unstimulated-

177 or stimulated-PBMCs supernatants. 24 hours after treatment, cell-free culture supernatants were

178 collected, aliquoted and stored at -80°C until further use. Cells were detached from the culture

179 plate using trypsin-EDTA (Sigma-Aldrich, St. Louis, MO) for downstream applications.

180

181 Immunostaining and flow cytometry

182 Astrocytes were immuno-stained for intracellular complement component 3 (C3) and serine

183 protease inhibitor family A member 3 (SERPINA3) (9,11). Briefly, cells were resuspended in

184 fixation/permeabilization solution (BD Cytofix/CytopermTM; BD Biosciences, San Jose, CA) for

185 20 min at 4ºC, washed with permeabilization/wash buffer (BD Perm/WashTM; BD Biosciences)

186 and stained with anti-C3 (Sigma-Aldrich; Catalog# GW20073F) or -SERPINA3 (R&D Systems;

187 Catalog# MAB1295) antibodies conjugated with fluorescence-tag (Lightning-Link PE-Cy7

188 Antibody Labeling Kit; Novus Biologicals, Centennial, CO). Stained cells were washed twice

189 and then analyzed using fluorescence-activated flow cytometry (BD LSR II Flow Cytometer, BD

190 Biosciences, San Jose, CA).

191

192 ELISA

193 C3 levels in astrocyte culture supernatants were measured using solid-phase sandwich ELISA

194 (Genway Biotech, San Diego, CA; Catalog# GWB-1C0767). All samples were diluted 1:1 with

195 bovine serum albumin (Sigma-Aldrich) and C3 concentrations were calculated using a standard

196 curve according to the manufacturer’s instructions.

197

198 Research subjects

199 All subjects were prospectively recruited under NIH IRB-approved protocol (Comprehensive

200 Multimodal Analysis of Neuroimmunological Diseases of the Central Nervous System,

201 ClinicalTrials.gov Identifier: NCT00794352), between 10/2008 and 04/2016. All subjects

202 underwent neurological examination to derive the measures of neurological disability Expanded

203 Disability Status Scale (EDSS) (12) and Combinatorial Weight-Adjusted Disability Scale

204 (CombiWISE) (13). Composite magnetic resonance imaging (MRI) scale of CNS tissue

205 destruction (COMRIS-CTD) was calculated from 3T brain MRI images as described (14). MS

206 severity measures were calculated based on published algorithms either from EDSS: Multiple

207 Sclerosis Severity Score (MSSS) (15) and Age-Related Multiple Sclerosis Severity (ARMSS)

208 (16) or from CombiWISE: Multiple Sclerosis Disease Severity Scale (MSDSS) (5). MS

209 diagnostic subgroups (RR-MS, secondary progressive MS [SP-MS] and primary progressive MS

210 [PP-MS]) were classified using McDonald’s criteria, 2010 and 2017 revisions (17).

211 Healthy donors (HD) and untreated MS patients (Table 1 and Supplementary Table 1) were

212 randomly divided into: training (n = 169) and validation cohort (n = 164), stratified on age,

213 gender, MS type and MS severity (10).

214

215 CSF collection and processing

216 CSF were collected by lumbar puncture and processed as per standard operating procedure (18).

217 CSF aliquots were prospectively labeled using alphanumeric code, and immediately stored on ice

218 after collection. Cell-free CSF supernatants were collected by centrifugation (335gx10 minutes at

219 4ºC), aliquoted, and stored at -80°C. Personnel processing CSF were blinded to diagnoses and

220 clinical outcomes.

221

222 DNA-aptamer-based multiplex proteomics

223 CSF and cell culture supernatants were analyzed using the Slow Off-rate Modified Aptamers

224 scan (SOMAscan; SomaLogic Inc., Boulder, CO) (19). CSF supernatants were analyzed using

225 the 1.1K SOMAscan platform (analyzes 1128 proteins, available from June 2012 to October

226 2016), and cell culture supernatants were analyzed using the 1.3K platform (analyzes 1317

227 proteins, available from October 2016).

228

229 Neurotoxicity

230 The human neuroblastoma cell line, SK-N-SH (ATCC® HTB-11TM; ATCC, Manassas, VA),

231 was cultured according to the manufacturer’s instruction. After 24 hours, cells were treated with

232 either unstimulated- or stimulated-astrocyte culture supernatants (50% v/v). After 24 hours

233 apoptotic neurons were analyzed using Annexin V-FITC (TACS® Annexin V Kit; Trevigen Inc.,

234 Gaithersburg, MD) as described (20).

235

236 Drug library screening

237 Astrocytes were plated (105 cells/ml) on poly-L-lysine (Sigma-Aldrich)-coated 96-well cell-

238 culture plates (100 µl/well). After 24 hours, cells were treated with either unstimulated- or

239 stimulated-PBMCs supernatants (50% v/v) in the presence of a respective drug (10 µM or 100

240 nM, 1431 drugs; Selleckchem LLC, Houston, TX; Catalog# L1300) or dimethyl sulfoxide

241 (DMSO, a drug solvent; Sigma-Aldrich; control). 24 hours after treatment, supernatants were

242 collected, and C3 levels were analyzed using ELISA. Percent change in absolute C3 secretion for

243 a drug treatment with respect to control was calculated: ([C3 secretion by stimulated astrocytes

244 with a drug treatment – C3 secretion by unstimulated astrocytes] / [C3 secretion by stimulated

245 astrocytes with DMSO treatment – C3 secretion by unstimulated astrocytes]) * 100. Cytotoxic

246 effects of respective drug treatments were analyzed using the 3-(4,5-dimethylthiazol-2-yl)-2,5-

247 diphenyltetrazolium bromide (MTT) assay (ThermoFisher Scientific, Waltham, MA) as per

248 manufacturer’s instructions.

249

250 Statistical analyses

251 To differentiate biomarkers specific for MS biology from physiological age- and gender-

252 differences, CSF SOMAscan values for all subjects were adjusted for age and gender

253 dependency within HD subgroup as described (21).

254 Group-wise comparisons were performed using Analysis of Variance (ANOVA). When

255 statistically significant (p < 0.05) differences were found, pairwise multiple comparisons using

256 Tukey’s p-value adjustments were performed. Kruskal-Wallis ANOVA examined differences

257 between diagnostic subgroups in biomarker values within the training cohort. When statistically

258 significant differences were found, pairwise multiple comparisons using Dunn’s p-value

259 adjustment were performed. Only statistically significant (adjusted p-value < 0.05) differences

260 were then validated in an independent validation cohort.

261 Relationship between astrocyte-specific biomarkers and clinical outcomes were examined using

262 Spearman correlations. Only statistically significant (p < 0.05) correlation coefficients in the

263 training cohort were then validated in an independent validation cohort.

264 Drugs from the library screen were grouped based on their known targets (153 groups). Mean C3

265 concentrations (normalized to control treatment) of drugs that did not induce substantial toxicity

266 (i.e., MTT > 75%), at 100 nM concentrations, were calculated and compared with C3 levels of

267 control (i.e. 100%) using ANOVA. P-values were adjusted for multiple comparisons using

268 Benjamini and Hochberg False Discovery Rate (FDR).

269

270

271

272

273

274 Results

275 Expression of toxic astrogliosis biomarkers

276 Induction of toxic astrogliosis by stimulated PBMCs was verified by demonstrating upregulation

277 of intracellular C3 and SERPINA3, known markers of toxic/reactive astrocytes (9–11,22)(Figure

278 1A) and by observing that toxic-astrocyte-conditioned medium (50% v/v) induces apoptosis of

279 neuronal cell line SK-N-SH (Figure 1B).

280

281 Identification of inflammation-induced, astrocyte-secreted biomarkers

282 Cell culture supernatants from unstimulated- and stimulated-astrocytes were collected before (0

283 h) and after (24 h) the respective treatments and analyzed using DNA aptamer-based proteomic

284 assay. Stimulation index for each protein was calculated by the ratio of relative fluorescence

285 units (RFU) under stimulated (24h / 0h) versus unstimulated (24h / 0h) conditions

286 (Supplementary Table 2). We arbitrarily defined measured proteins as inflammation-induced,

287 astrocyte-secreted biomarkers if their stimulation index was > 5.

288 18 proteins were identified as inflammation-induced, astrocyte-secreted biomarkers: C-X-C

289 motif chemokine ligand (CXCL-6, 9, 10 and 11), matrix metalloproteinases (MMP-3, 10, 12 and

290 13), C-C motif chemokine ligand (CCL-7, 8 and 20), SERPINA3, C-X3-C motif chemokine

291 ligand 1 (CX3CL1), TNFα induced protein 6 (TNFAIP6), C3, complement factors (CFB and

292 CFH) and interleukin 1 receptor like 1 (IL1RL1).

293

294 Analysis of biomarkers across disease diagnosis subgroups

295 Age- and gender-adjusted SOMAscan values for inflammation-induced astrocyte-secreted

296 biomarkers were compared among diagnostic subgroups (HD, RR-MS, P-MS [comprised of both

297 SP-MS and PP-MS]) in the training cohort. Statistically significant differences were validated in

298 an independent validation cohort. CXCL10 and C3 were significantly elevated only in RR-MS

299 patients compared to HD, and MMP13 was elevated only in P-MS patients. While, SERPINA3

300 was significantly elevated in both MS subgroups (RR-MS and P-MS) (Figure 2A and

301 Supplementary Table 3).

302

303 Correlation analysis between biomarkers and clinical outcome measures

304 To determine how inflammation-induced, astrocyte-secreted biomarkers change with MS

305 progression, RFU values for inflammation-induced, astrocyte-secreted biomarkers were

306 correlated with the disability outcomes EDSS (12), CombiWISE (13), and with MRI scale of

307 CNS tissue destruction (COMRIS-CTD) (14). No statistically significant correlations were

308 observed (Supplementary Table 4).

309 To assess the potential pathogenic role of toxic astrocyte-secreted biomarkers in MS, we

310 correlated biomarker RFUs (adjusted for natural aging and gender effects as described in

311 Methods) with validated measures of MS severity: MSSS (15), ARMSS (16), and MSDSS (5).

312 Only SERPINA3 had reproducible correlations with two MS severity outcomes, ARMSS (ρ =

313 0.19 and adjusted p = 0.0229) and MSDSS (ρ = 0.21 and adjusted p = 0.0147) (Figure 2B) in the

314 independent validation cohort. This suggests that CSF SERPINA3 levels reflect the pathogenic

315 role of toxic astrocytes in MS-associated CNS tissue destruction.

316

317 Drug library screening

318 We applied a commercially available drug library to in vitro model of inflammation-induced

319 astrocytes to identify therapeutic targets to impede induction of the “toxic astrocyte” phenotype.

320 Efficacy of respective drugs was analyzed by astrocyte-driven secretion of C3, a marker used to

321 identify toxic astrocytes in MS lesions (9,11).

322 Drugs were first tested at 10 μM, but at this concentration, many drugs induced >75% cytotoxic

323 effects on astrocytes (i.e., MTT < 75% of that of the control treatment; Supplementary Table 5).

324 Thus, the entire screen was repeated with 100-fold lower drug concentrations (i.e., 100 nM),

325 which better reflects in vivo achievable concentrations of tested drugs in humans (Figure 3).

326

327 Identification of signaling pathways important in inflammation-induced transformation of

328 astrocytes towards a toxic phenotype

329 For pathway analysis, drugs with common therapeutic target that were not cytotoxic (i.e.,

330 MTT<75% of control) were grouped together. For these groups we calculated mean % inhibition

331 of C3 secretion and formally tested the statistically significant group’s effects. The full results

332 are in Supplementary Table 6, while Figure 3 provides results for relevant drug categories.

333 Inhibitors of multiple pathways that interact together and participate in endoplasmic reticulum

334 (ER) stress response (i.e., unfolder protein response; UPR) reached formal statistical significance

335 in this analysis. Specifically, inhibitors of the following targets (arranged in the order of

336 descending potency) reduced C3 secretion from inflammation-induced astrocytes by at least

337 25%: 1) proteasome, 2) heat shock protein 90 (HSP90), 3) p38 mitogen-activated protein kinase

338 (p38 MAPK), 4) phosphoinositide 3-kinases (PI3K), 5) mammalian target of rapamycin

339 (mTOR), 6) Akt, and 7) sphingosine 1-phosphate (S1P) receptor. Additionally, dantrolene, an

340 anti-spasticity drug that inhibits calcium release from the ER of CNS cells via inhibiting

341 ryanodine receptor (RyR), also reliably inhibited secretion of C3 with high potency in the drug

342 library screen (i.e., 77.7 % inhibition). Inhibitors of c-kit and poly ADP ribose polymerase

343 (PARP) also achieved statistical significance (FDR-adjusted p <0.05), although their inhibitory

344 effect did not reach 75%. The summary of signaling pathways found inhibitory in our screen and

345 their physiological relationship is depicted in Figure 4.

346 Surprisingly, most glucocorticoids induced secretion of C3 (174.44 % of control) from

347 inflammation-induced astrocytes. Although other drug groups also stimulated C3 secretion from

348 toxic astrocytes with formal statistical significance, the group effects were lower than 125%.

349 These stimulatory drug targets include cytochrome P450s (CYPs), adrenergic receptor,

350 cyclooxygenase (COX), leukotriene receptor (LTR), and anti-infection agents. Drugs that did not

351 pertain to any specific category and thus were grouped as “others” also achieved a minimal

352 stimulatory effect that reached formal statistical significance (Figure 3).

353

354 Concentration-response curves for selected drugs

355 Because the drug library screen was performed only in two concentrations (10 μM and 100 nM),

356 the representative drugs from the most effective target categories and other drugs of potential

357 interest were validated in independent experiments, using a concentration curve of 0, 10, 100,

358 and 1,000 nM concentrations. 50% inhibitory concentrations (IC50) were calculated from these

359 curves (Figure 5).

360 All tested proteasome inhibitors (Delanzomib, Ixazomib, Carfilzomib and Marizomib) blocked

361 C3 secretion from inflammation-induced astrocytes at low concentrations (IC50 < 25 nM). Out of

362 three tested HSP90 inhibitors, Ganetespib had the highest potency (IC50 = 29.75 nM), while the

363 other two drugs, Tanespimycin and Onalespib, had an IC50 193.8 and 120.9 nM, respectively.

364 Within tested mTOR-inhibitors, Rapamycin had lowest IC50 (159.7 nM), and Everolimus and

365 Tacrolimus had IC50 918.6 and 2888 nM respectively; While, based on tested concentrations,

366 IC50 for Temsirolimus cannot be determined. Dantrolene, an RyR-antagonist, had projected IC50

367 226.5 nM.

368

369 Effects of selected drugs on inflammation-induced astrocyte-mediated neuronal apoptosis

370 To validate that inhibition of C3 secretion from toxic astrocytes also inhibits neuronal apoptosis,

371 we tested the effects of selected drugs on toxic astrocyte-induced neuronal apoptosis in vitro.

372 SK-N-SH cells were treated with astrocyte-conditioned medium (treated with unstimulated or

373 stimulated PBMC supernatants in the presence of a respective drug or DMSO; 50% v/v), and

374 neuronal apoptosis was analyzed 24 hours later. Treatment with supernatants from inflammation-

375 induced astrocytes significantly enhanced neuronal apoptosis compared to supernatants from

376 unstimulated astrocytes (Figure 6). Most tested drugs prevented inflammation-induced astrocyte-

377 mediated neuronal apoptosis, except Delanzomib, Ixazomib and Ganetespib. Delanzomib

378 treatment significantly elevated neuronal apoptosis, suggesting its direct neurotoxic effects.

379

380 Proteomic analyses of supernatants from inflammation-induced astrocytes in the presence of

381 selected drugs

382 While we performed drug library screening using astrocyte-secretion of C3, because this marker

383 has been widely used as evidence of a toxic astrocyte signature in pathology studies (9), C3 is

384 unlikely to be directly neurotoxic. Proteomic analysis identified multiple biomarkers secreted by

385 inflammation-induced astrocytes, out of which only SERPINA3 validated significant relationship

386 to MS severity in an independent validation cohort (Figure 2).

387 Therefore, to understand the efficacy of drugs that inhibit C3 secretion on the proteome of

388 inflammation-induced astrocytes, we selected three representative drugs (Ganetespib [HSP90-

389 inhibitor], Dantrolene [RyR-antagonist], and Rapamycin [mTOR-inhibitor]) and studied their

390 influence on the secretome of inflammation-induced astrocytes using SOMAscan. Ganetespib

391 was selected based on its reproducibly strong efficacy on C3 secretion, while Dantrolene and

392 Rapamycin were studied for their potential use as candidate inhibitors of toxic astrocytes in

393 proof-of-principle clinical trial in MS.

394 Ganetespib strongly reduced secretion of all biomarkers previously identified as part of a “toxic

395 astrocyte signature” (Figure 7). However, Ganetespib also inhibited secretion of proteins from

396 astrocytes that were not activated by inflammatory stimuli (Supplementary Table 7). This

397 suggest inhibition of physiological functions of astrocytes, which may have detrimental effects

398 on neurons in vivo. Dantrolene reduced secretion of most of the “toxic astrocyte” biomarkers,

399 including SERPINA3. However, it also elevated secretion of MMPs (MMP-10 and 12). In

400 contrast, Rapamycin did not have potent effect on secretion of most of the “toxic astrocyte”

401 biomarkers, except CX3CL1, C3, MMP13 and IL1RL1. Rapamycin elevated secretion of

402 CCL20, TNFAIP6 and CFB.

403

404

405

406

407

408

409

410

411

412 Discussion

413 There are two approaches to drug library screening: in vitro assays and in vivo experiments in

414 short-lived animal species, such as fruit-flies or zebrafish (23,24). Both have limitations: the non-

415 physiological approach of in vitro assays risks the possibility that obtained results that do not

416 reproduce the in vivo situation, while animal models suffer from differences in physiological and

417 pathogenic mechanisms between lower species and humans. Nevertheless, drugs identified

418 through in vitro drug screens validated therapeutic efficacy in humans (25).

419 We acknowledge following drawbacks of our study: 1) In vitro monoculture might have

420 influenced the astrocyte phenotype, as was previously observed on transcriptome level (26); 2)

421 The short-term induction of toxic astrocytes by inflammatory stimuli may not capture spectrum

422 of toxic astrocytes in vivo, where long-term exposure to an inflammatory environment may cause

423 epigenetic changes not reversible by pharmacological manipulations; 3) While we selected

424 inhibition of C3 for drug library screen as C3 is broadly used to identify toxic astrocytes in

425 human brain, including MS lesions (9), the mechanism(s) of neurotoxicity by toxic astrocytes

426 have not been elucidated and are unlikely caused by C3. Therefore, inhibition of C3 does not

427 guarantee inhibition of astrocyte-induced neurotoxicity. Thus, every drug screen requires

428 validation through interventional clinical trial(s) to prove or disprove that the targeted

429 mechanism was truly pathogenic.

430 Being mindful of the limitations, we employed strategies to maximize the clinical relevance of

431 drug library screen: we analyzed overlap between the toxic astrocyte-secreted proteome in our in

432 vitro model and previously published studies and found overlap for CXCLs (11), SERPINA3

433 (9,11,27), C3 (9), CFB (9) and MMPs (28). This suggests that our assay captures most of the in

434 vivo phenotypical change of ‘toxic astrogliosis’.

435 C3 has been widely used as a marker of the A1/toxic astrocytes (9,29,30) and Liddelow et al

436 found C3 protein mainly expressed in astrocytes (9). However, under neuroinflammatory

437 conditions, CSF C3 concentrations cannot be solely attributed to astrocytes because on an

438 mRNA level, C3 is mainly expressed in microglia and immune cells of myeloid lineage

439 (Supplementary Table 8). This is aligned with our observation that CSF C3 is significantly

440 elevated only in RR-MS. This suggests that during the formation of MS lesions, most C3 either

441 originates from serum and reaches CSF due to blood-brain barrier (BBB) opening or is secreted

442 by cells of the myeloid lineage recruited to acute MS lesions (2,3). Thus, while CSF C3 cannot

443 be a reliable biomarker of toxic astrogliosis in vivo, the same problem is not associated with in

444 vitro astrocyte monocultures, used in library screen.

445 From all inflammation-induced astrocyte-secreted proteins, only SERPINA3 CSF levels are

446 astrocyte-specific, as infiltrating immune cell or other CNS cells has minimal SEPINA3 mRNA

447 expression (Supplementary Table 8). The relevance of the CSF SERPINA3 as a biomarker of

448 toxic astrocytes is supported by animal studies. Genomic analysis of astrocytes from pro-

449 inflammatory stimuli-treated mice has shown robust increase in SERPINA3 (11). SERPINA3

450 also induced toxic effects on cortical murine neuron cultures (27), suggesting its direct

451 pathogenic role. Whether this pathogenicity is true in humans requires clinical trial evidence.

452 Nevertheless, we conclude that CSF SERPINA3 is the best biomarker for in vivo

453 pharmacodynamic readout of the inhibitory effect therapies on toxic astrogliosis.

454 The first important insight from our study identifies small molecules among FDA-approved MS

455 drugs with inhibitory effect on toxic astrocytes. One study reported that in vitro pre-treatment of

456 astrocytes with dimethyl fumarate (DMF; 25 µM) reduced secretion of proinflammatory

457 cytokines and oxidative stress following stimulation with IL1β (31). We too observed that DMF

458 inhibited C3 secretion by astrocytes at a non-physiologically high (10 µM) concentration but not

459 at 100 nM. A 25 µM dose is almost 1000-fold higher than the peak measured concentration in

460 human blood (32) and the concentrations in the CNS are likely even lower. This highlights the

461 essential limitation of testing inhibitory effects with non-physiological drug concentrations. In

462 contrast, we based our conclusions on 100 nM drug concentration screen and validated the most

463 promising drugs in dose titration experiments (down to 10 nM).

464 In contrast to DMF, our drug library screen identified S1P receptor modulators as inhibiting C3

465 secretion from toxic astrocytes. Indeed, immunostaining of postmortem MS brains showed

466 elevated expression of S1P receptors on astrocytes in MS lesions (33), suggesting a role of S1P

467 receptors in the induction of toxic astrogliosis. Additionally, pretreatment of human induced

468 pluripotent stem cells (iPSCs)-differentiated astrocytes with fingolimod or siponimod reduced

469 secretion of proinflammatory cytokines in the presence of inflammatory stimuli via blocking

470 activation and nuclear translocation of the NFκB-p65 (34). Our results expand on these data and

471 suggest that S1P receptor modulators may also partially block the process of toxic astrogliosis in

472 vivo. We plan to test this hypothesis in future studies evaluating the effect of S1P receptor

473 modulators on CSF SERPINA3 levels.

474 While, glucocorticoids, effective inhibitors of neuro-inflammation (35,36), reproducibly elevated

475 secretion of C3 from inflammation-induced astrocytes in vitro, there is currently no experimental

476 evidence that increased secretion of C3 captures the complete neurotoxic phenotype of

477 astrocytes. Thus, it will be important to analyze CSF SERPINA3 levels in patients treated with

478 steroids in future studies.

479 Next, we’ll provide integrative analysis of signaling pathways that library screen identified as

480 contributing to formation of toxic astrocytes (please refer to Figure 4). Among these, HSP and

481 proteasome inhibitors were the most potent. In response to inflammation, astrocytes initiate

482 protein biosynthesis to amplify the inflammatory response. Most of these synthesized proteins

483 are secreted, requiring posttranslational modifications and proper folding in the ER before

484 entering the Golgi apparatus. High protein secretion may overwhelm protein folding capacity,

485 resulting in accumulation of unfolded proteins, known as proteotoxic or ER stress. This triggers

486 UPR response aimed to restore homeostasis, consisting of the following actions/pathways: 1)

487 refolding of misfolded proteins, 2) degradation of irreparably damaged proteins via the ubiquitin

488 proteasome system (UPS) and lysosome/autophagy-mediated pathways, and 3) downregulation

489 of new protein biosynthesis (37,38). The UPR is initially protective, but under sustained stress, it

490 induces further inflammation and eventually triggers apoptosis (39).

491 By inhibiting protein folding and degradation of misfolded proteins, HSP and proteasome

492 inhibitors exacerbate ER stress and stall protein synthesis. Indeed, Ganetespib, an HSP-inhibitor,

493 reduced secretion of all astrocyte-secreted proteins, both physiological and inflammation-

494 induced. Clearly, such a strategy is not sustainable long-term, as it will likely cease essential

495 astrocyte-mediated functions that may not be consistent with the survival. Even in oncology,

496 these drugs can be administered only short-term. Thus, we excluded these drugs/pathways from

497 consideration for treatment of neurodegenerative diseases.

498 Another class of drugs identified through our drug library screening are PI3K/Akt/mTOR-

499 inhibitors. This pathway is known to positively regulate the proinflammatory response via

500 upregulating activation and nuclear translocation of NFκB-p65 (40,41). Particularly mTOR

501 signaling plays a vital role in astrocytic proliferation and production of proinflammatory

502 mediators during stress (42). mTOR has a bidirectional crosstalk with ER stress: 1) mTOR

503 signaling works upstream and exacerbates ER stress via upregulating protein biosynthesis and

504 downregulating clearance of damaged proteins through the lysosome/autophagy-mediated

505 pathways, and 2) mTOR signaling also works downstream of ER stress, where sustained ER

506 stress activates mTOR signaling, leading to inflammation and cell death (41,43,44). Though our

507 results showed that PI3K/Akt/mTOR-inhibitors inhibit secretion of C3 from inflammation-

508 induced astrocytes, Rapamycin failed to inhibit secretion of SERPINA3, and actually increased

509 secretion of some pro-inflammatory molecules, such as CCL20, CCL8. This is an intriguing

510 observation as it suggests that different pathways mediate secretion of C3 versus other markers

511 of toxic astrocytes, like SERPINA3. Because the reviewed literature, supported by observed

512 correlation of CSF SERPINA3 with MS severity assigns a stronger role to SERPINA3 than C3

513 as the marker of pathogenic astrocytes, PI3K/Akt/mTOR inhibitors may not inhibit the most

514 relevant neurotoxic functions.

515 Dantrolene, an RyR-antagonist, significantly and reproducibly reduced secretion of C3, but also

516 SERPINA3, CCL20, CCL8 from inflammation-induced astrocytes. Under normal conditions, the

517 ER has at least a three-fold higher Ca2+ concentration than that of the cytosol, which is crucial

518 for normal protein folding. Early ER stress dysregulates RyR functioning (45), causing Ca2+ leak

519 from the ER, which further disturbs normal protein folding. During sustained ER stress,

520 dysregulated RyR-mediated Ca2+ leak from the ER positively regulates UPR-mediated

521 inflammation and cell death (45,46), suggesting that dantrolene should alleviate ER stress. This

522 would mean that drugs with opposing mechanisms on ER stress (i.e., HSP/proteasome inhibitor

523 exacerbating and dantrolene alleviating ER stress) are both efficacious in suppressing generation

524 of toxic astrocytes. How can we resolve this discrepancy? One possibility is that the neurotoxic

525 astrocytes are not induced directly by inflammatory stimuli, which signal via PI3K/Akt/mTOR

526 and p38MAPK. Instead, the result of this pro-inflammatory signaling is robust protein synthesis

527 and ER stress. Perhaps it is the failing ER stress phase of the astrocytic response to inflammation

528 that generates their true neurotoxic phenotype. The HSP/proteasome inhibition tips the astrocyte

529 one way (shutting off protein synthesis completely and eventually causing cell death), while

530 dantrolene blocks the pathogenic transformation of astrocytes by blocking Ca2+ release from the

531 failing ER to cytosol and/or to mitochondria. Clearly, this hypothesis will need to be tested in

532 future studies. Nevertheless, since dantrolene is FDA-approved for treatment of spasticity and

533 has been applied in this indication to MS patients long-term, it is an immediately available

534 candidate for testing in proof-of-principle clinical trials its efficacy on toxic astrocyte inhibition.

535 Its major drawback is serious hepatotoxicity with doses over 400 mg/day (47).

536 In conclusion, this study elucidated signaling pathways associated with transformation of

537 astrocytes to a toxic phenotype and identified candidate drugs for clinical testing. The first step

538 in proving efficacy of any of these agents should be demonstrating their pharmacodynamic effect

539 on CSF SERPINA3 levels, which correlate with MS severity.

540

541 Acknowledgments

542 We thank Elena Romm for processing of CSF samples. We thank Dr. Alison Wichman and

543 research nurses Mary Sandford and Tiffany Hauser for expert patient care and patient care

544 coordinator Michelle Woodland for patient scheduling. Finally, we thank all the patients, their

545 caregivers and healthy volunteers, without whom this work could not be possible.

546

547

548

549

550 References

551 1. Markowitz CE. Multiple sclerosis update. Int. J. MS Care 2014;16:5–11.

552 2. Frischer JM, Bramow S, Dal-Bianco A, et al. The relation between inflammation and

553 neurodegeneration in multiple sclerosis brains. Brain 2009;132(Pt 5):1175–89.

554 3. Kirk J, Plumb J, Mirakhur M, McQuaid S. Tight junctional abnormality in multiple

555 sclerosis white matter affects all calibres of vessel and is associated with blood-brain

556 barrier leakage and active demyelination. J. Pathol. 2003;201(2):319–27.

557 4. Komori M, Blake A, Greenwood M, et al. Cerebrospinal fluid markers reveal intrathecal

558 inflammation in progressive multiple sclerosis. Ann. Neurol. 2015;78(1):3–20.

559 5. Weideman AM, Barbour C, Tapia-Maltos MA, et al. New Multiple Sclerosis Disease

560 Severity Scale Predicts Future Accumulation of Disability. Front. Neurol.

561 2017;8(NOV):598.

562 6. Yi W, Schlüter D, Wang X. Astrocytes in multiple sclerosis and experimental

563 autoimmune encephalomyelitis: Star-shaped cells illuminating the darkness of CNS

564 autoimmunity. Brain. Behav. Immun. 2019;80(March):10–24.

565 7. Kamermans A, Planting KE, Jalink K, et al. Reactive astrocytes in multiple sclerosis

566 impair neuronal outgrowth through TRPM7-mediated chondroitin sulfate proteoglycan

567 production. Glia 2019;67(1):68–77.

568 8. Brambilla R. The contribution of astrocytes to the neuroinflammatory response in multiple

569 sclerosis and experimental autoimmune encephalomyelitis. Acta Neuropathol.

570 2019;137(5):757–783.

571 9. Liddelow SA, Guttenplan KA, Clarke LE, et al. Neurotoxic reactive astrocytes are

572 induced by activated microglia. Nature 2017;541(7638):481–487.

573 10. Masvekar R, Wu T, Kosa P, et al. Cerebrospinal fluid biomarkers link toxic astrogliosis

574 and microglial activation to multiple sclerosis severity. Mult. Scler. Relat. Disord.

575 2019;28:34–43.

576 11. Zamanian JL, Xu L, Foo LC, et al. Genomic analysis of reactive astrogliosis. J. Neurosci.

577 2012;32(18):6391–6410.

578 12. Kurtzke JF. Rating neurologic impairment in multiple sclerosis: An expanded disability

579 status scale (EDSS). Neurology 1983;33(11):1444–1452.

580 13. Kosa P, Ghazali D, Tanigawa M, et al. Development of a Sensitive Outcome for

581 Economical Drug Screening for Progressive Multiple Sclerosis Treatment. Front. Neurol.

582 2016;7(AUG):131.

583 14. Kosa P, Komori M, Waters R, et al. Novel composite MRI scale correlates highly with

584 disability in multiple sclerosis patients. Mult. Scler. Relat. Disord. 2015;4(6):526–35.

585 15. Roxburgh RHSR, Seaman SR, Masterman T, et al. Multiple sclerosis severity score:

586 Using disability and disease duration to rate disease severity. Neurology

587 2005;64(7):1144–1151.

588 16. Manouchehrinia A, Westerlind H, Kingwell E, et al. Age Related Multiple Sclerosis

589 Severity Score: Disability ranked by age. Mult. Scler. 2017;23(14):1938–1946.

590 17. Polman CH, Reingold SC, Edan G, et al. Diagnostic criteria for multiple sclerosis: 2005

591 Revisions to the “McDonald Criteria”. Ann. Neurol. 2005;58(6):840–846.

592 18. Barbour C, Kosa P, Komori M, et al. Molecular-based diagnosis of multiple sclerosis and

593 its progressive stage. Ann. Neurol. 2017;82(5):795–812.

594 19. Gold L, Ayers D, Bertino J, et al. Aptamer-based multiplexed proteomic technology for

595 biomarker discovery. PLoS One 2010;5(12):e15004.

596 20. Masvekar R, Mizrahi J, Park J, et al. Quantifications of CSF Apoptotic Bodies Do Not

597 Provide Clinical Value in Multiple Sclerosis. Front. Neurol. 2019;10:1241.

598 21. Barbour C, Kosa P, Varosanec M, et al. Molecular models of multiple sclerosis severity

599 identify heterogeneity of pathogenic mechanisms. medRxiv 2020;2020.05.18.20105932.

600 22. Catts VS, Wong J, Fillman SG, et al. Increased expression of astrocyte markers in

601 schizophrenia: Association with neuroinflammation. Aust. N. Z. J. Psychiatry

602 2014;48(8):722–734.

603 23. MacRae CA, Peterson RT. Zebrafish as tools for drug discovery. Nat. Rev. Drug Discov.

604 2015;14(10):721–731.

605 24. Pandey UB, Nichols CD. Human disease models in drosophila melanogaster and the role

606 of the fly in therapeutic drug discovery. Pharmacol. Rev. 2011;63(2):411–436.

607 25. Mei F, Fancy SPJ, Shen YAA, et al. Micropillar arrays as a high-throughput screening

608 platform for therapeutics in multiple sclerosis. Nat. Med. 2014;20(8):954–960.

609 26. Wilhelm A, Volknandt W, Langer D, et al. Localization of SNARE proteins and secretory

610 organelle proteins in astrocytes in vitro and in situ. Neurosci. Res. 2004;48(3):249–257.

611 27. Padmanabhan J, Levy M, Dickson DW, Potter H. Alpha1-antichymotrypsin, an

612 inflammatory protein overexpressed in Alzheimer’s disease brain, induces tau

613 phosphorylation in neurons. Brain 2006;129(11):3020–3034.

614 28. Hara M, Kobayakawa K, Ohkawa Y, et al. Interaction of reactive astrocytes with type i

615 collagen induces astrocytic scar formation through the integrin-N-cadherin pathway after

616 spinal cord injury. Nat. Med. 2017;23(7):818–828.

617 29. Xu X, Zhang A, Zhu Y, et al. MFG-E8 reverses microglial-induced neurotoxic astrocyte

618 (A1) via NF-κB and PI3K-Akt pathways. J. Cell. Physiol. 2018;234(1):904–914.

619 30. Hyvärinen T, Hagman S, Ristola M, et al. Co-stimulation with IL-1β and TNF-α induces

620 an inflammatory reactive astrocyte phenotype with neurosupportive characteristics in a

621 human pluripotent stem cell model system. Sci. Rep. 2019;9(1)

622 31. Galloway DA, Williams JB, Moore CS. Effects of fumarates on inflammatory human

623 astrocyte responses and oligodendrocyte differentiation. Ann. Clin. Transl. Neurol.

624 2017;4(6):381–391.

625 32. Linker RA, Haghikia A. Dimethyl fumarate in multiple sclerosis: latest developments,

626 evidence and place in therapy. Ther. Adv. Chronic Dis. 2016;7(4):198–207.

627 33. Van Doorn R, Van Horssen J, Verzijl D, et al. Sphingosine 1-phosphate receptor 1 and 3

628 are upregulated in multiple sclerosis lesions. Glia 2010;58(12):1465–1476.

629 34. Colombo E, Bassani C, De Angelis A, et al. Siponimod (BAF312) Activates Nrf2 While

630 Hampering NFκB in Human Astrocytes, and Protects From Astrocyte-Induced

631 Neurodegeneration. Front. Immunol. 2020;11(April):1–11.

632 35. Rhen T, Cidlowski JA. Antiinflammatory action of glucocorticoids - New mechanisms for

633 old drugs. N. Engl. J. Med. 2005;353(16):1711–1723.

634 36. Giatti S, Boraso M, Melcangi RC, Viviani B. Neuroactive steroids, their metabolites, and

635 neuroinflammation. J. Mol. Endocrinol. 2012;49(3)

636 37. Bozaykut P, Ozer NK, Karademir B. Regulation of protein turnover by heat shock

637 proteins. Free Radic. Biol. Med. 2014;77:195–209.

638 38. Schröder M, Kaufman RJ. ER stress and the unfolded protein response. Mutat. Res. -

639 Fundam. Mol. Mech. Mutagen. 2005;569(1–2):29–63.

640 39. Hotamisligil GS. Endoplasmic Reticulum Stress and the Inflammatory Basis of Metabolic

641 Disease. Cell 2010;140(6):900–917.

642 40. Gao S, Liu W, Zhuo X, et al. The activation of mTOR is required for monocyte pro-

643 inflammatory response in patients with coronary artery disease. Clin. Sci.

644 2015;128(8):517–526.

645 41. Bové J, Martínez-Vicente M, Vila M. Fighting neurodegeneration with rapamycin:

646 Mechanistic insights. Nat. Rev. Neurosci. 2011;12(8):437–452.

647 42. Li CY, Li X, Liu SF, et al. Inhibition of mTOR pathway restrains astrocyte proliferation,

648 migration and production of inflammatory mediators after oxygen-glucose deprivation and

649 reoxygenation. Neurochem. Int. 2015;83–84:9–18.

650 43. Appenzeller-Herzog C, Hall MN. Bidirectional crosstalk between endoplasmic reticulum

651 stress and mTOR signaling. Trends Cell Biol. 2012;22(5):274–282.

652 44. Kato H, Nakajima S, Saito Y, et al. MTORC1 serves ER stress-triggered apoptosis via

653 selective activation of the IRE1-JNK pathway. Cell Death Differ. 2012;19(2):310–320.

654 45. Yamamoto WR, Bone RN, Sohn P, et al. Endoplasmic reticulum stress alters ryanodine

655 receptor function in the murine pancreatic cell. J. Biol. Chem. 2019;294(1):168–181.

656 46. Wang Y, Shi Y, Wei H. Calcium Dysregulation in Alzheimer’s Disease: A Target for

657 New Drug Development. J. Alzheimer’s Dis. Park. 2017;7(5)

658 47. Utili R, Boitnott JK, Zimmerman HJ. Dantrolene-Associated Hepatic Injury: Incidence

659 and character. Gastroenterology 1977;72(4):610–616.

660

661

662

663

664

665 Figures

666

667 Figure 1: (A) Representative flow cytometry images of astrocytes immuno-stained for C3 and

668 SERPINA3, 24 hours after treatment with unstimulated- or stimulated-PBMCs supernatants.

669 Normalized, intracellular expression (mean fluorescence intensity) of C3 and SERPINA3 under

670 different culture conditions is represented in respective charts. Expression across treatment

671 groups (n = 3) were compared using ANOVA (Tukey's multiple comparisons test); *p < 0.05,

672 ***p < 0.0005 and ****p < 0.0001. (B) Representative flow cytometry images of neuroblastoma

673 cell line (SK-N-SH) stained with Annexin V after treatment with unstimulated- or stimulated-

674 astrocyte conditioned media (Astrocyte supernatant) for 24 hours. Normalized % of apoptotic

675 (Annexin V+) neurons, compared (n = 4) using paired t test; *p < 0.05.

676

677

678 Figure 2: (A) Age- and gender-adjusted SOMAscan values (procedure defined unit, p.d.u.) of

679 inflammation-induced, astrocyte-secreted biomarkers were compared across disease diagnostic

680 subgroups using Kruskal-Wallis ANOVA (Dunn’s multiple comparisons test). *p < 0.05 and **p

681 < 0.005. The ‘+’ sign represents mean of respective subgroup and dotted line indicates the

682 median of the HD subgroup. (B) Correlations between SOMAscan values and clinical outcome

683 measures were analyzed using Spearman analyses. The solid line indicates the best-fit line for

684 linear regression between respective variables and the dotted line represents the 95% confidence

685 interval. Spearman’s rank correlation coefficient (ρ) and adjusted P values are represented on

686 respective correlation plots. Only statistically significant and reproducible results from the

687 validation cohort are presented.

688

689

690 Figure 3: Toxicity (MTT) and efficacy of each drug (at 100 nM) in blocking secretion of C3 was analyzed. Effect of each drug was

691 normalized with respect to control (DMSO) treatment: ([C3 secretion by stimulated astrocytes with a drug treatment – C3 secretion by

692 unstimulated astrocytes] / [C3 secretion by stimulated astrocytes with DMSO treatment – C3 secretion by unstimulated astrocytes]) *

693 100. 1431 drugs (n = 1) are represented here; significantly efficacious drugs at group level (mean C3 < 75% of control and adjusted p-

694 value < 0.05) are represented with respective colors.

695

696 Figure 4: The summary of signaling pathways found inhibitory in our screen and their

697 physiological relationships. Our results suggest that NFκB-p65 mediated transcription of

698 proinflammatory mediators plays a central role in induction of toxic astrogliosis. Abnormally

699 high protein secretion during inflammatory stress may overwhelm the protein folding capacity of

700 the ER, causing accumulation of unfolded proteins, leading to disturbed Ca2+-homeostasis and

701 proteotoxic stress. Sustained ER stress further exacerbates the toxic astrocyte-mediated

702 proinflammatory response and under severe stress, it may lead to cell death. Drugs which can

703 alleviate NFκB-mediated proinflammatory responses and relieve ER stress may be effective in

704 blocking astrocyte-mediated toxicity during MS progression. Green color and ‘+’ sign indicate

705 positive regulation. Red color and ‘stop’ sign indicate negative regulation.

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725 Figure 5: Concentration-response curves for selected drugs. Efficacy of blocking C3 secretion of proteasome-, HSP90-, and mTOR-

726 inhibitors, and dantrolene were tested at 0, 10, 100, and 1000 nM concentrations. Drugs were tested in triplicates for each

727 concentration. Mean ± standard deviations are represented here. The dotted line indicates a 50% reduction in C3 secretion, compared

728 to control, and IC50 for each drug are depicted on respective graphs. For better representation, y-axis limits are selected to be from 0

729 to 100%, but some datapoints and error bars may be out of axis limits.

730

731 Figure 6: (A) Absolute (24h – 0h) secretion of C3 by astrocytes under different culture

732 conditions were analyzed using ELISA. The effect of each tested drug (100 nM) was normalized

733 with respect to control (DMSO) treatment: ([C3 secretion by inflammation-induced astrocytes

734 with a drug treatment – C3 secretion by unstimulated astrocytes] / [C3 secretion by

735 inflammation-induced astrocytes with DMSO treatment – C3 secretion by unstimulated

736 astrocytes]) * 100. (B) Unstimulated or stimulated ACM-mediated neuronal apoptosis was

737 analyzed using Annexin V immunostaining and flow cytometry; Normalized % of apoptotic

738 (Annexin V+) neurons. C3 secretion and neuronal apoptosis across different culture conditions

739 (n = 3) were compared using ANOVA (Tukey's multiple comparisons test). *p < 0.05 vs

740 unstimulated + DMSO and #p < 0.05 vs. inflammation-induced + DMSO.

741

742 Figure 7: Normalized stimulation indices for inflammation-induced astrocytes secreted biomarkers after treatment with Ganetespib,

743 Dantrolene and Rapamycin. Stimulation index for each protein, under DMSO or respective drug treatment, was calculated by taking

744 the ratio of SOMAscan values under inflammation-induced culture conditions versus unstimulated culture conditions. Stimulation

745 indices for each drug were then normalized with respect to DMSO treatment (stimulation index [% control]).

746

747

748

749

750

751 Tables

Training Cohort (N = 169) Validation Cohort (N = 164) Diagnosis HD RR-MS P-MS P value HD RR-MS P-MS P value N (female/male) 8/10 35/23 44/49 0.2431 10/11 35/19 48/41 0.2965 Average 39.9 40.2 52.6 <0.0001 35.6 39.5 53.9 <0.0001 Age (SD) 15.6 11.3 10.4 11.0 10.6 9.0 range 19.4 - 70.3 18.0 - 68.7 22.0 - 65.8 19.7 - 57.4 18.3 - 67.9 29.8 - 74.7 Average - 2.0 5.5 <0.0001 - 1.6 5.3 <0.0001 EDSS (SD) - 1.5 1.6 - 1.0 1.5 range - 0.0 - 6.5 1.5 - 8.5 - 0.0 - 6.0 2.0 - 7.5 Average - 15.6 43.0 <0.0001 - 12.5 42.2 <0.0001 CombiWISE (SD) - 10.8 14.7 - 6.9 13.6 range - 2.2 - 50.4 9.4 - 84.5 - 2.4 - 35.5 14.5 - 70.8 Average - 8.4 15.0 <0.0001 - 8.4 14.6 <0.0001 COMRIS-CTD (SD) - 5.7 6.2 - 5.5 5.9 range - 0.0 - 22.0 3.7 - 31.5 - 0.0 - 23.5 0.0 - 26.3 Average - 4.0 6.8 <0.0001 - 3.6 6.8 <0.0001 MSSS (SD) - 2.2 2.0 - 2.2 1.8 range - 0.2 - 9.2 1.2 - 10.0 - 0.2 - 9.4 1.3 - 9.8 Average - 3.5 6.5 <0.0001 - 3.1 6.2 <0.0001 ARMSSS (SD) - 2.3 2.2 - 1.9 2.2 range - 0.2 - 9.5 0.9 - 9.9 - 0.3 - 7.4 0.9 - 9.5 Average - 1.4 2.1 <0.0001 - 1.3 2.1 <0.0001 MSDSS (SD) - 0.7 1.1 - 0.5 0.9 range - 0.4 - 3.5 0.3 - 5.3 - 0.5 - 2.6 0.7 - 4.4 752

753 Table 1: All subjects were divided into two cohorts, training and validation. Gender distribution

754 (Chi-square test), age (ANOVA), and clinical outcome measures (EDSS, CombiWISE,

755 COMRIS-CTD, MSSS, ARMSS and MSDSS; t-test) were compared across diagnostic categories

756 (HD, RR-MS and P-MS [comprised of both SP-MS and PP-MS]).

757

758

759

760

761

762

763

764 Supplementary Table Legends

765 Supplementary Table 1: Cohort (training- and validation-cohort), diagnosis (HD, RRMS,

766 SPMS and PPMS), clinical outcome measures (EDSS, CombiWISE, COMRIS-CTD, MSSS,

767 ARMSS and MSDSS) and age- and gender-adjusted CSF SOMAscan (1.1K platform; only toxic

768 astrocyte-specific markers, 17 proteins) data of all subjects (n = 333). Patients were recoded and

769 personal identification information (PII, such as age, gender and clinic visit dates) were

770 excluded.

771

772 Supplementary Table 2: SOMAscan (1.3K platform, 1317 proteins) data of astrocyte culture

773 supernatants, under different culture conditions (unstimulated and stimulated), at 0 and 24 hours.

774 Stimulation index (stimulated [24h / 0h] / unstimulated [24h /0h]) for each protein were

775 calculated.

776

777 Supplementary Table 3: Within training cohort, differences for inflammation-induced,

778 astrocytes-secreted biomarkers’ SOMAscan values (age- and gender-adjusted) across disease

779 diagnostic subgroups (HD, RR-MS and P-MS) were analyzed using Kruskal-Wallis ANOVA

780 (Dunn’s multiple comparisons test). Only statistically significant (adjusted p-value < 0.05)

781 differences were then validated in an independent validation cohort.

782

783 Supplementary Table 4: Correlations between age- and gender-adjusted SOMAscan values for

784 inflammation-induced, astrocytes-secreted biomarkers and clinical outcome measures were

785 analyzed using Spearman analysis. Only statistically significant (adjusted p-value < 0.05)

786 correlations were then validated in an independent validation cohort. Table represents

787 Spearman’s rank correlation coefficient (ρ) and P values.

788

789 Supplementary Table 5: Drug library (Selleckchem; FDA-approved Drug Library) screening

790 data. All drugs (1431) were first screened at 10 μM, and then at 100 nM. Toxicity (MTT assay)

791 and efficacy (inhibiting absolute secretion of C3) of each drug were analyzed. MTT and C3

792 assay results for each plate were normalized based on respective control treatment (stimulated +

793 DMSO, % control). To summarize overall effect of each drug at two different concentrations,

794 mean and standard deviation (SD) were calculated.

795

796 Supplementary Table 6: Drugs with common therapeutic targets were grouped together (1431

797 drugs were grouped into 151 groups). At 100 nM concentration, only for safe drugs (MTT >

798 75% of control), mean C3 concentrations (% of control) for each group were calculated and

799 compared with control (DMSO; C3 = 100%), p-values were adjusted for multiple comparisons.

800

801 Supplementary Table 7: SOMAscan (1.3K platform, 1317 proteins) data of astrocyte culture

802 supernatants, under different culture conditions: unstimulated and inflammation-induced, in

803 presence of DMSO or selected drugs (Ganetespib, Dantrolene and Rapamycin; 100 nM), 0 and

804 24 hours. Stimulation indices for each protein, under DMSO or respective drug treatment, were

805 calculated. Then stimulation indices for each drug were normalized with respect to DMSO

806 treatment (% control).

807

808 Supplementary Table 8: Relative expression (RNA-Seq analyses) of inflammation-induced,

809 astrocytes-secreted biomarkers, within human CNS and blood cells. Data from publicly available

810 databases were extracted and compiled. Human CNS cells database:

811 https://www.brainrnaseq.org/; Human blood cells database:

812 https://www.proteinatlas.org/about/download.

813